optimum

85376e33 - Distribute and complete onnxruntime tests (decoder models) (#2278)

Distribute and complete onnxruntime tests (decoder models) (#2278) * added test_decoders.py * fix position ids for single batch and more complete decoder testing for finegrained debugging * support merging seq2seq models when used as decoders and add more tests * fix pipe tests * update phi min transformers version (broken by cache position refactorization) * remove deprecated bloom modeling * update opt onnx config to the one with position ids * remove all complex deprecated modeling * get_supported_model_type_for_task should only return suooprted model types * update min transformers * use transformers like api for use_cache and add can_use_cache and is_merged * testing * fix * fix * remove unnecessary * simply qwen3 * docs * qwen-moe * model type shenanigans * fix * use test models from optimum-internal-hf with proper metadata * Update optimum/onnxruntime/modeling_decoder.py Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com> * keep supported model types * optimum model * fix failing test by forcing export --------- Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>